Camoufox CLI

The evidence clearly describes the daemon, Unix socket, sessions, tabs, cookie import/export, persistent identities, proxies, local files, upstream Camoufox attribution, and authorship. Deductions apply because the tool can navigate, click, fill forms, execute arbitrary page JavaScript, read or write cookies, save screenshots, and access networks through proxies without confirmations, domain restrictions, or privilege tiers. Proxy credentials may appear in plaintext command lines and configuration, with no documented permission or redaction controls for configuration, cookies, profiles, or sockets. Only coupled dependencies are exactly pinned; Pillow and development dependencies are not fully locked, and browser acquisition can fall back to scraping release pages without documented checksums, signatures, audits, or vulnerability response. Closing sessions and manually deleting profiles provide limited recovery, but remote web actions are generally irreversible. No concrete malware, credential theft, covert exfiltration, or other red-line evidence is present.

The README, package metadata, and tests are broadly consistent about core commands, configuration precedence, session isolation, tab state, and error paths. There are targeted tests for concurrent recovery and the last-tab-close race. Deductions reflect the absence of implementation files, recorded test results, and end-to-end evidence, plus the README command catalog omitting the tested pdf command. Installation requirements, tightly coupled dependency pins, and the GitHub rate-limit fallback are documented, but browser releases, system packages, and dual npm/pip distribution chains remain availability dependencies. Missing arguments, unknown actions, an unlaunched browser, and malformed configuration have error or warning evidence, although not every failure branch asserts a specific actionable message.

The intended audiences and scenarios are thoroughly described for AI agents and developers, including browsing, extraction, parallel tabs, multiple identities, and multi-account isolation, so audience and scenarios receive full credit. Commands, flags, configuration precedence, and persistent versus per-launch state are reasonably bounded, but acceptable-use, site-policy, dangerous-action, and anti-detection boundaries are not documented. CLI triggers are precise and parameter validation is tested; however, the actual Agent Skill files are absent, preventing a complete assessment of natural-language activation and false-trigger avoidance. npm, pipx, Linux dependencies, minimum Node/Python versions, and configuration paths are covered, while macOS, Windows, containers, permissions, and parity between distributions are not comprehensively addressed.

The README has strong information architecture and complete installation, quick-start, command, configuration, persistence, and architecture sections, earning full installation credit. Naming is mostly stable and many commands are tested, but pdf is absent from the README and no Node/Python release-synchronization mechanism is shown. Examples are extensive, though there is no dedicated FAQ or systematic troubleshooting guide. GitHub rate limits, shared-browser queueing, unenforced tab-name uniqueness, and Alpha status are disclosed, but anti-detection limitations, compatibility risks, and legal or site-terms concerns are not consolidated. The MIT license is consistent across README, metadata, and the complete LICENSE, justifying full marks. Version 0.7.3 and Alpha status are present, but there is no changelog or upgrade policy. An author and issue tracker are identifiable, yet publisher identity is unverified and no maintenance policy, support commitment, or security-reporting route is supplied.

Accessibility snapshots, stable references, JSON output, text, screenshots, PDF, cookies, and tab/session results are designed for direct shell and agent consumption, justifying full output-usability credit. Camoufox integration, persistent fingerprints, and parallel isolation offer clear incremental value over a basic browser CLI. The central claims of no detectable Playwright leakage and effective anti-detection lack static comparative evidence, detection results, or included implementation support, so marginal value is not fully demonstrated. Installation is concise and daemon/browser sharing can reduce repeated work, but browser downloads, system dependencies, proxy management, persistent state, and operational or compliance burdens keep cost-benefit below comprehensive treatment.

Command behavior, defaults, configuration precedence, errors, and concurrency semantics are traceable across documentation, metadata, and multiple test files. Deductions apply because fingerprint spoofing, GeoIP behavior, persistent identity implementation, and the claim of no bot-detectable Playwright leaks lack corresponding implementation files, focused tests, or measured results in the supplied evidence. The supplied sources partially corroborate version, license, commands, and configuration, but all originate from the same repository and include neither execution records nor independent corroboration. Documentation does not clearly distinguish tested facts, intended design, and marketing-level effectiveness claims, so fact/inference separation is weak.